# ContextAtlas

Context infrastructure for AI coding agents that combines tree-sitter code chunking, LanceDB vector search and SQLite full-text search into hybrid retrieval and persistent project memory, exposed as a CLI, MCP server or library.

ContextAtlas is open-source context infrastructure for AI coding agents, providing hybrid code retrieval, project memory and retrieval observability as a CLI, an MCP server, or an embeddable library. It combines tree-sitter semantic chunking, LanceDB vector search, SQLite FTS5 full-text search and token-aware context packing to deliver structured code context to tools such as Claude Code, Codex and custom agent workflows.

It aims to answer whether an agent can find the right code faster in a large repository, whether repository understanding can persist across sessions instead of being rediscovered each time, and whether retrieval, indexing and memory quality can be observed and improved over time. Recent changes include git-hook-driven automatic index maintenance with a fast health check, extraction of Memory MCP tools' business logic into a separate application layer, an embedding gateway with local caching and multi-upstream routing including Hugging Face, and churn- and cost-aware index planning.

It is written in TypeScript and MIT licensed.

Website: https://github.com/codefromkarl/ContextAtlas